Saturday, July 1, 2017

Does this data look like a downward spiral, or just the bottom of a ski hill?

Some people normally think visually. They can easily imagine what a graph will look like. Others may need to learn to think that way, and to use software like Microsoft Excel or PowerPoint to turn a dull data table into an informative graph.   

On June 26 the Daily Mail had an article by Chris Spargo titled Megyn Kelly ratings dip for third straight week hitting new season low of 3.41M viewers on NBC as host announces she will be off the air next Sunday during limited summer run. But when Jane Genova blogged about it that day she instead claimed Megyn Kelly’s ‘Sunday Night’ – Ratings in Downward Spiral.

That show had led off with a highly-advertised episode. It was highlighted by an interview with Russian president Vladimir Putin. Then from the first to the second week viewers dropped from 6.1 million to 3.61 million, or by a horrible 41%. But from the second week to the third, with 3.50 million, the drop was just 3%. And from the third week to the fourth, with 3.41 million, the drop again was by just 3%. When you graph those numbers, you will see Jane was being overly dramatic. It’s just the bottom of a ski hill.  

Megyn Kelly’s show still had fewer viewers than either the 7.21 million for 60 Minutes, and 3.92 million for America’s Funniest Home Videos. But 60 Minutes has been on for almost five decades. Going against it with a similar news show almost is a suicide mission.

Another way to analyze those numbers is by looking at differences and slopes. We can ask a simple question - when will the show have no viewers left? After two weeks things looked dire – like no one would be watching by week four. But that was not what happened.

In my first career as a research metallurgist I reviewed magazine articles submitted for publication in both a metallurgy magazine and a corrosion magazine, I learned to carefully check whether the data meant what the authors claimed. On March 18, 2013 I blogged about What is your hearing threshold? – the joy of statistics. There I discussed how the same statistics also apply to stress corrosion cracking (SCC) tests. Once I read a submitted article where the highest stress level for SCC tests didn’t make the specimens crack. The author wrongly assumed that if they just had gone one step higher they would have. That’s wishful thinking, not engineering, so I rejected it. (D. J. Finney’s book on Statistical Method in Biological Assay warns researchers NOT to ever make that assumption).

No comments: